llm.ClassificationLLM

llm.ClassificationLLM(self, model_name=config['llm']['llm_model_name'], llm=None, embedding_handler=None, max_tokens=1600, temperature=0.0, verbose=False, openai_api_key=None)

Wraps the logic for using an LLM to classify respondent’s data based on provided index. Includes direct (one-shot) generative llm method and Retrieval Augmented Generation (RAG).

Parameters

Name	Type	Description	Default
model_name	str	Name of the model. Defaults to the value in the config file. Used if no LLM object is passed.	config['llm']['llm_model_name']
llm	LLM	LLM to use. Optional.	None
embedding_handler	EmbeddingHandler	Embedding handler. Optional. If None a default embedding handler is retrieved based on config file.	None
max_tokens	int	Maximum number of tokens to generate. Defaults to 1600.	1600
temperature	float	Temperature of the LLM model. Defaults to 0.0.	0.0
verbose	bool	Whether to print verbose output. Defaults to False.	False
openai_api_key	str	OpenAI API key. Optional, but needed for OpenAI models.	None

Methods

Name	Description
get_sic_code	Generates a SIC classification based on respondent’s data
get_soc_code	Generates a SOC classification based on respondent’s data
rag_general_code	Generates a classification answer based on respondent’s data
rag_sic_code	Generates a SIC classification based on respondent’s data using RAG approach.

get_sic_code

llm.ClassificationLLM.get_sic_code(industry_descr, job_title, job_description)

Generates a SIC classification based on respondent’s data using a whole condensed index embedded in the query.

Parameters

Name	Type	Description	Default
industry_descr	str	Description of the industry.	required
job_title	str	Title of the job.	required
job_description	str	Description of the job.	required

Returns

Type	Description
SicResponse	Generated response to the query.

get_soc_code

llm.ClassificationLLM.get_soc_code(job_title, job_description, level_of_education, manage_others, industry_descr)

Generates a SOC classification based on respondent’s data using a whole condensed index embedded in the query.

Parameters

Name	Type	Description	Default
job_title	str	The title of the job.	required
job_description	str	The description of the job.	required
level_of_education	str	The level of education required for the job.	required
manage_others	bool	Indicates whether the job involves managing others.	required
industry_descr	str	The description of the industry.	required

Returns

Type	Description
SocResponse	The generated response to the query.

Raises

Type	Description
ValueError	If there is an error parsing the response from the LLM model.

rag_general_code

llm.ClassificationLLM.rag_general_code(respondent_data, candidates_limit=7)

Generates a classification answer based on respondent’s data using RAG and custom index.

Parameters

Name	Type	Description	Default
respondent_data	dict	A dictionary containing respondent data.	required
candidates_limit	int	The maximum number of candidate codes to consider. Defaults to 7.	7

Returns

Type	Description
RagResponse	The generated classification response to the query.

Raises

Type	Description
ValueError	If there is an error during the parsing of the response.
ValueError	If the default embedding handler is required but not loaded correctly.

rag_sic_code

llm.ClassificationLLM.rag_sic_code(industry_descr, job_title=None, job_description=None, expand_search_terms=True, code_digits=5, candidates_limit=5)

Generates a SIC classification based on respondent’s data using RAG approach.

Parameters

Name	Type	Description	Default
industry_descr	str	The description of the industry.	required
job_title	str	The job title. Defaults to None.	None
job_description	str	The job description. Defaults to None.	None
expand_search_terms	bool	Whether to expand the search terms to include job title and description. Defaults to True.	True
code_digits	int	The number of digits in the generated SIC code. Defaults to 5.	5
candidates_limit	int	The maximum number of SIC code candidates to consider. Defaults to 5.	5

Returns

Type	Description
SicResponse	The generated response to the query.

Raises

Type	Description
ValueError	If there is an error during the parsing of the response.
ValueError	If the default embedding handler is required but not loaded correctly.